The Agent Skills Directory

[COMMAND_EXECUTION]: The skill provides templates and instructions for creating scripts that execute shell commands using execFileSync in Node.js. It details how these scripts are registered in hooks.json to be executed by the agent platform during specific events.
[DYNAMIC_EXECUTION]: The documentation covers the creation and runtime execution of custom scripts (Node.js, Python, Shell) as plugin hooks. It also describes type: prompt and type: agent hooks, which involve runtime evaluation of instructions and tool-using agents to verify tasks.
[INDIRECT_PROMPT_INJECTION]: The skill describes a platform surface for processing data from the workspace via hooks.
Ingestion points: Hook scripts ingest untrusted tool inputs, tool outputs, and user prompts via process.stdin (documented in SKILL.md).
Boundary markers: The guide recommends structured JSON responses ({"ok": true}) and explicit reasoning for blocking actions.
Capability inventory: Hooks have access to the file system, network (via type: http hooks or custom scripts), and subprocess execution (execFileSync).
Sanitization: The documentation explicitly recommends security best practices, including quoting variable expansions, checking for path traversal (..), and skipping sensitive file paths like .env or .ssh (documented in the 'Security Rules' section of SKILL.md).
[EXTERNAL_DOWNLOADS]: The skill references official documentation hosted on code.claude.com. This is a well-known service associated with the agent's ecosystem.

hook-creator